Hi all, I am trying to implement an application similar to Nest camera's familiar face feature. Specifically, it works like this:
- The 1st ever detected face will be a new one and we need to give a name to it.
- From the next detected face, the app needs to decide whether it is similar to one of the stored face. If not it is a new one and we need to give a new name for this person.
So I think it will involve 2 important tasks:
- Unknown face classification
- Active learning, so that once a new face is added, the application can quickly get a new classifier.
I've been ignoring the active learning part, i.e., a new classifier is trained at every step of getting a new face, but still the classification is pretty poor.
Any ideas or suggestions would be greatly appreciated!
Thanks